Ever used AI at work but didn't tell your boss? That instinct to hide it might be the real barrier to getting good at it.

This explores what managers do that pushes employees from casual AI use toward deep, skilled use, or holds them back. The corpus doesn't have a study that follows workers through that journey. What it does have points to a surprising answer: the biggest thing a manager may shape isn't access or training. It's whether using AI feels safe to admit.

Start with the hidden cost. Across four experiments with more than 4,000 people, Do people fear judgment when they use AI at work? found that AI users expect to be seen as less competent and less hardworking. As a result, they are less willing to tell managers and colleagues that they use it. That matters because nobody becomes a skilled practitioner in secret. If people hide their AI use, they can't share techniques, ask for better tools, or get credit for new workflows. Their use stays private and shallow.

This suggests why manager support seems to matter. Gallup's 2026 survey, covered in Does manager support actually shape how employees experience AI at work?, found that employees whose managers actively back AI use report an improved workplace culture at nearly double the rate of others (31% vs. 21%). Read next to the Reif findings, one plausible way this works is that visible approval from the manager removes the fear of being judged. Once AI use is something you're expected to do, there's no reason to hide it, and openly experimenting is how people build skill. The data is correlational, though. Teams that are already engaged may simply have better managers and also rate their culture higher.

The other side of the picture is what heavy users look like. Does delegating work to AI actually damage worker skills? reports that people who hand the most work to Claude are also the most optimistic about their careers and feel their skills are gaining value. That fits a self-reinforcing loop: people who feel free to delegate heavily come to believe it pays off, and then lean in further. But the sample is Anthropic's own users, so it shows a link, not cause and effect. Meanwhile, Where have workers actually delegated tasks to AI? shows that deep delegation, where AI tasks become part of how work is structured, is concentrated in information-heavy jobs and follows what the technology can actually do. Manager behavior probably matters most there, where the capability already exists and the remaining barrier is social and organizational.

To be direct about the gaps: most of what the search returned is about "frontier" AI models (scheming, reward hacking, evaluation awareness), not frontier practitioners. The collection has no direct evidence on specific manager actions such as training budgets, setting an example, or protected time to experiment. What it does suggest is a question worth carrying forward: the most useful thing a manager can do may be to make AI use visible and respected, because hidden use rarely turns into skilled use.